[nycbug-talk] ZFS question

[George Rosamond]

On 09/26/11 22:36, Charles Sprickman wrote:
>
> On Sep 26, 2011, at 9:50 PM, Isaac Levy wrote:
>
>> On Sep 26, 2011, at 9:13 PM, George Rosamond wrote:
>>
>>> Quick question for the ZFS users in the audience:

Wow.  Thanks.

The box is an old 64-bit box and it has 8 x 750 gig.  Probably going to
runFreeNAS, and some data will be just writes (remote backups) and 
others will be more read-intensive (local network NFS).  Probably around 
6 gig of RAM for now.

Let me snip and paste *some* of the relevant points.  All contributions 
were appreciated.

> Cool- I did this a while back when ZFS on FreeBSD was really green,
> I hacked hw together with as many sata cards as I could dig out- ZFS
> was itself rough-edged and new back then so my results were, um, a
> good learning experience :)

Was looking for my other controllers. .. that explains it!

> Sounds like a tall order to test, but it's so easy- since there's no
> waiting around for newfs to complete...  And if it bombs out, you'll
> have a chance to see why.

Will probably do some of this for that and other reasons.

> I'd love to see some vendor donate a bunch of different disks, a few
> decent multi-disk chassis, and setup some real tests... but until
> there's some real benchmarks all I can back that up with is my .02¢
>
> (Anyone know how we could pull off NYC*BUG collaboration with Tom's
> Hardware or something?)

This could be interesting.  On the pure advocacy level, benchmark
testing (even showing updated ZFS versions on FBSD or NBSD) would be
worthwhile.

Anyone interested in doing this ping Ike offline :)

Or certainly raise it at the next NYCBUG meeting or on a separate thread
here.

> I always recommend to setup a hotspare (or 2) with any zpool. I've run into
> data errors in the past with zfs, but you can usually fix it with a zpool

Very much. . . on the same page here.


> This one sometimes gives interesting results depending on what type
> of data you're storing.  Hint: Being able to read more than one block
> worth of data while reading just one block off the drive can speed
> some things up and the cpu hit is negligible.
>

Understood.  It's really quite a simple role. . . some will likely be 
www data, but while the served media files can be large, it shouldn't be 
an issue.  It's not some psycho session or intensive db data.

And yes, I have followed the FBSD wiki stuff for a long while, including 
the loader.conf stuff.

Funny stuff Spork.  This would be a fun in-meeting demonstration if 
planned well.

But you have worked almost every first Wednesday since we started in 
2004. . .what can we do? ;)

Good stuff.

g